Detecting Intentions of Vulnerable Road Users Based onCollective Intelligence as a Basis for Automated Driving

点击翻译按钮获取中文摘要

The present renewal proposal of the project DeCoInt2 focuses on intention detection of vulnerable road users (VRUs) in automated driving using cooperative technologies. Especially in urban areas, VRUs such as pedestrians and cyclists will still play an important role in the mixed traffic of tomorrow. For an accident-free traffic with automated vehicles, it is not just important to perceive VRUs but it is also essential that their intentions are detected. The intention detection consists of basic movement detection, e.g., standing, moving, turning, and a forecast of the future trajectory. We aim to further contribute to our envisioned future traffic scenarios, in which the collective intelligence of vehicles equipped with sensors such as cameras and radar, maps, and Car-2X communication capabilities, sensor-equipped infrastructure, and VRUs themselves (if equipped with smart devices such as smartphones) are used to detect potentially dangerous situations involving VRUs, before the affected entities face these situations. In the first phase, our contribution to that envisioned future traffic scenario was cooperative VRU intention detection including cooperative perception, basic movement detection, and trajectory forecast. We managed to realize a robust tracking of VRUs, resolving occlusions by means of multiple traffic participants including smart devices worn by VRUs in a cooperative way. Using our cooperative approach, we achieved fast and robust basic movement detection even in situations with limited visibility or bad environmental conditions, enabling an accurate VRU trajectory forecast. In the second phase of the project DeCoInt2, we will consider the following three new aspects: identification and integration of context information, probabilistic VRU trajectory forecasts, and situation analysis and prediction. Additionally, we will extend our work on cooperative tracking, basic movement forecasting, and integration of smart devices. We aim to integrate context information (e.g., geometric information such as a course of a bicycle path or traffic regulations) to improve the intention detection process, allowing for more precise forecasts. The combined results of the cooperative perception incorporating context information, cooperative basic movement forecasts, and cooperative probabilistic trajectory forecasts, which take the form of predictive probability distributions, serve as an important ingredient for situation analysis and prediction, which in turn is essential for trajectory planning. Altogether, we aim to evaluate our cooperative approach in selected sample scenarios conducted on-line in a real traffic environment.